Nvidia waits for US approval to ship H200 chips to China

Nvidia is waiting for approval from the US government to ship its H200 chips to China, according to the company’s CFO.

The chipmaker has seen high demand from China after President Donald Trump reversed a longstanding ban on sending the chips there last year.

The US government is currently reviewing license applications, but there is no confirmed timeline for when shipments can begin.

Nvidia did not comment on any talks with Chinese officials about the licenses.

Nvidia has already discussed future data center expansions with customers for 2027, but has not provided sales guidance.

The company recently introduced six new chips under its next-generation “Vera Rubin” line, which are now in full production.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

Nvidia is waiting for Commerce Department clearance to ship H200s to China 1. The Export Control Reform Act (ECRA), a U.S. law on export controls, gives national security staff and Congress oversight 1.
H200s exceed the Total Processing Performance (TPP) ceiling by about 10x 2. Trump cleared H200 exports in December 2025 with a 25% fee, yet approval is not assured, which could dent Nvidia’s path to $500 billion in sales by end-2026 1.
Shipping H200s would raise China’s advanced AI compute, which lags the U.S. due to manufacturing bottlenecks 2. That shift could shrink the U.S. 21-49x compute edge (the relative amount of AI processing power available) to as low as 1.3x in worst cases 2.

Nvidia is discussing 2027 data center buildouts as global power could grow from 103 GW to 200 GW by 2030, with $3 trillion over five years including $1.2 trillion in real estate 3.
Power vendors can target hyperscalers (very large cloud providers operating massive data centers) building 2,000+ MW sites 4. Some campuses could draw 5 GW, enough for five million homes 4. That scale opens work in generation, transmission, and grid interconnections 4.
Cooling vendors should prioritize liquid and immersion 5. AI servers need 50-100 kW per rack versus 5-7 kW 5. Air cooling is impractical above 20 kW, so retrofits are needed for next-gen chips 5.